AC / DC hybrid grid planning method based on climate response capability risk analysis

Abstract

The invention discloses an AC / DC hybrid grid planning method based on climate response capability risk analysis, and the method comprises the steps: constructing a time sequence generation model based on a denoising diffusion Transform, capturing a multi-time scale nonlinear coupling relation between new energy output data and meteorological data in a planning region through a diffusion process and a cross-scale attention mechanism, and carrying out the prediction of the multi-time scale nonlinear coupling relation. Generating a combined sample set; extracting a risk feature vector of the new energy output scene from the joint sample set, and establishing a mixed Gaussian distribution representation risk feature space; based on an outlier detection algorithm, identifying a low probability-high influence scene from the risk feature space to form a typical day set for planning; based on the typical day set for planning, constructing a mixed integer planning model with minimization of line investment cost, operation cost and various penalty costs as targets; and solving the mixed integer programming model to obtain an alternating-current and direct-current hybrid grid planning scheme.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system planning, and more particularly to a method for AC-DC hybrid network planning based on climate response capability risk analysis. BACKGROUND

[0002] High proportion of new energy access makes AC-DC hybrid power grid face new challenges of significantly enhanced climate dependence. Traditional network planning methods are mostly based on the static probability distribution of historical meteorological data, which is difficult to capture the temporal and spatial differentiation characteristics of renewable energy output caused by long-term evolution of climate system, resulting in structural risks of planning scheme in response to extreme climate events. Especially under the background of global warming, extreme weather such as typhoon and drought occurs frequently, and the coupling relationship between clean energy output such as wind power and photovoltaic and climate elements presents significant nonlinear and time-varying characteristics. The existing scene generation method represented by typical day is obviously insufficient in climate adaptability on the interannual scale. How to establish a correlation model between the multi-scale physical process of climate system and the long-period characteristics of power grid planning has become a core problem to improve the resilience of new energy power system.

[0003] The current new energy power system planning faces three technical bottlenecks: firstly, in the aspect of climate-new energy coupling modeling, the traditional statistical method cannot simultaneously represent the combined influence of short-term weather fluctuations and long-term climate pattern evolution, resulting in insufficient spatio-temporal fidelity of output scene; secondly, the risk scene screening link lacks an active capture mechanism for low-probability-high-impact events, and existing clustering algorithms cannot quantitatively measure key characteristic indicators such as climate response vulnerability entropy and spatio-temporal risk dispersion; thirdly, the climate risk conduction effect is not fully considered in AC-DC coordinated planning, and the utilization rate constraint of DC line is separated from the evaluation of dynamic capacity expansion potential, causing the mismatch of power transmission resources across space-time scales. These problems seriously restrict the long-term economy and operation reliability of new power system under the carbon neutralization target. SUMMARY

[0004] Therefore, the present application provides a method for AC-DC hybrid network planning based on climate response capability risk analysis, which at least solves some of the technical problems in the background art.

[0005] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0006] A method for AC-DC hybrid network planning based on climate response capability risk analysis, comprising the following steps:

[0007] A time series generation model based on denoising diffusion Transformer is constructed, and through diffusion process and cross-scale attention mechanism, the multi-time scale nonlinear coupling relationship between new energy output data and meteorological data in the planning area is captured, and a joint sample set is generated;

[0008] Extracting risk feature vectors of new energy output scenarios from the joint sample set and establishing a mixed Gaussian distribution to represent the risk feature space;

[0009] Based on the outlier detection algorithm, low probability and high impact scenarios are identified from the risk feature space to form a set of typical days for planning;

[0010] Based on the typical day set for planning, a mixed integer programming model is constructed with the goal of minimizing the line construction cost, operation cost and various penalty costs;

[0011] The mixed integer programming model is solved to obtain an AC / DC hybrid grid planning scheme.

[0012] Furthermore, the steps of constructing a time series generation model based on the denoising diffusion transformer include:

[0013] Constructing a diffusion process model of climate-new energy coupling relationship:

[0014]

[0015] where q(·) represents the transition probability of the diffusion process; X0 and Y0 represent the original climate variables and the new energy output sequence, respectively; n and Ω N Respectively represent the index and set of diffusion steps, the total number of diffusion steps is recorded as N; X n and Y n represent the climate variables and renewable energy output sequence at the nth step in the diffusion process; Indicates the mean The covariance matrix is ​​β n The multivariate Gaussian distribution function of I is used in the diffusion model to describe the noise injection process from step n-1 to step n; β n is the noise scheduling coefficient of the nth step, satisfying β1<β2<…<β N ; Indicates the previous step data (X n-1 ,Y n-1 ) scaling; I is the unit matrix, indicating that each dimension is independently noised, and the corresponding β n I represents the diagonal covariance matrix, which means that the variance added independently in each dimension is β n Gaussian noise; x t and y t They represent the original climate variables and the new energy output values ​​at time t respectively; T is the number of sequence periods; d1 and d2 represent the vector dimensions of the original climate variables and the new energy output respectively; and denote the real vector space of d1 dimension and d2 dimension respectively;

[0016] The climate variable sequence X injected with diffusion noise n and new energy output sequence Y n , which are mapped to the query matrix, key matrix and value matrix in the multi-head attention mechanism, and are trained through the projection weight W i Q 、W i K and W i V Feature transformation is performed, and finally multiple attention heads calculate the correlation patterns of different subspaces in parallel. The output matrix integrates the multi-head information to form a hidden state representation with physical meaning.

[0017] Furthermore, the multi-time-scale nonlinear coupling relationship between renewable energy output data and meteorological data in the planning area is captured to generate a joint sample set, which specifically includes the following steps:

[0018] The hidden state representation with physical meaning is gradually de-noised through the inverse process, and the multi-time-scale nonlinear coupling relationship between the renewable energy output data and meteorological data in the planning area is captured.

[0019] Furthermore, the hidden state representation with physical meaning is gradually de-noised through the reverse process, which is specifically achieved through the following formula:

[0020]

[0021] where p θ represents the transition probability of the inverse denoising process; μ θ (·) and Σ θ (·) are the mean and covariance functions parameterized by the Transformer neural network, respectively, used to predict the distribution of denoised data; α n It represents the cumulative noise attenuation coefficient considering the overall influence of the noise in the first n steps; h n is the hidden state generated by the cross-scale attention mechanism; P t is the temporal position encoding; T(·) represents the cross-scale attention mechanism; ε θ (·) represents the Transformer noise prediction network; diag(·) represents the function of converting a vector into a diagonal matrix; exp(·) represents the exponential function; represents the logarithmic variance prediction vector of the conditional Transformer network output; β n is the noise scheduling coefficient of the nth step; β s is the noise scheduling coefficient of the sth step.

[0022] Furthermore, the risk feature vector of the new energy output scenario is extracted from the joint sample set, and a mixed Gaussian distribution is established to represent the risk feature space, specifically including:

[0023] The risk feature vectors of the new energy output scenarios are extracted from the joint sample set, and a mixed Gaussian distribution is established to represent the risk feature space, specifically including:

[0024] The denoising diffusion transformer temporal generation model generates K renewable energy output scenarios through the transition probability of the reverse denoising process;

[0025] Extract risk feature vectors from the K new energy output scenarios, and each scenario s k Corresponding to a risk feature vector, a mixed Gaussian distribution is established based on the risk feature vector to represent the risk feature space:

[0026]

[0027] Specifically, the variables defined in formula (10) are shown in formulas (11)-(17):

[0028]

[0029]

[0030] in It's a scene k Normalized output sequence of new energy stations under ; Represents scene s k The distribution function of the system power shortage under ; express α quantile of ; is the expected operation function; j represents the index of the new energy station; N Re Indicates the total number of new energy stations; For scenes k The normalized output of the new energy station j under ; For scenes k Average output of new energy stations under Me / l Me and Ω Me Represents the index and set of climate variables respectively; m Me is the climate variable dimension; and Represents scene s k Climate variables under Me and l Me The vulnerability coefficient is expressed as the partial derivative of formula (10); It's a scene k Climate variables under Me Normalized influence weight of n Wa / l Wa and Ω Waare the index and set of wavelet decomposition scales respectively; N Wa is the total number of scales of wavelet decomposition; It's a scene k Under time scale n Wa Wavelet transform coefficients on ; is the scale n Wa energy share.

[0031] Furthermore, based on the outlier detection algorithm, low-probability-high-impact scenarios are identified from the risk feature space to form a set of typical days for planning, specifically including:

[0032] Assume that the feature space obeys Q G A mixture of Gaussian distributions, as shown in formula (18),

[0033]

[0034] where q G is the index of the Gaussian distribution; For the qth G The mixing weight of the components is such that the sum of the components is 1; and The qth G The mean vector and covariance matrix of the components;

[0035] The outlier scene set is defined as the samples that fall outside the α-confidence interval of all Gaussian distributions, that is, the outlier point determination is shown in formula (19):

[0036]

[0037] Among them L LQ is the outlier scene set; 2 is the critical value of the chi-square distribution, and thus the set of new energy output scenarios that are judged to be low probability-high impact scenarios is obtained;

[0038] The new energy output scenario set of low probability-high impact scenario and the conventional climate-new energy scenario set sampled by the inverse transition probability of the diffusion process are combined to form the typical day set for planning.

[0039] Furthermore, based on the typical day set for planning, a mixed integer programming model is constructed with the goal of minimizing line construction costs, operating costs, and various penalty costs. Specifically, the model includes:

[0040]

[0041] Among them C inv It is the sum of the investment and construction and system operation costs in the year; C ACinv and C DCinvare the investment and construction costs of AC and DC lines respectively; m / Ω rep and t / Ω T Represents the index / aggregation of typical day and intraday time periods respectively; p m represents the corresponding probability of a typical day m; and They represent the power shortage penalty cost, PV power curtailment penalty cost, wind power curtailment penalty cost, thermal power unit startup cost, thermal power unit shutdown cost, and thermal power unit operation cost in the t-th period of typical day m, respectively; represents the penalty cost of hydropower curtailment in a typical day m.

[0042] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method for planning AC / DC hybrid grids based on climate responsiveness risk analysis, which has the following beneficial effects:

[0043] (1) This paper constructs a new energy climate response modeling method based on a denoising diffusion transformer, which achieves an organic integration of the multi-scale physical processes of the climate system and the probability distribution of new energy output. While retaining the characteristics of short-term fluctuations, this method can capture the output heterogeneity caused by climate evolution, providing a reliable input basis for subsequent risk analysis. In particular, through the parameterized Gaussian transfer in the conditional denoising process, the model can adaptively correct the climate response deviation at different time scales.

[0044] (2) This paper proposes a hierarchical scenario screening mechanism driven by risk sensitivity. By constructing a risk feature space containing multidimensional features such as conditional risk value, spatiotemporal risk dispersion, and vulnerability entropy, and using mixed Gaussian distribution for outlier detection, it achieves accurate capture of extreme risk scenarios caused by the nonlinear evolution of the climate system.

[0045] (3) This paper constructs an AC / DC hybrid grid planning model based on the risk analysis of the long-term climate responsiveness of renewable energy. By embedding a set of risk scenarios into a set of typical daily scenarios, a mixed integer linear programming model is constructed that includes line construction costs, multi-timescale operating costs, and risk penalty costs. By designing constraints, the model can take into account the differentiated utilization characteristics of AC / DC lines and the operating requirements of multiple types of units. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0047] Figure 1 The present invention provides a schematic diagram of the overall process of the method. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] The embodiment of the present invention discloses a method for planning AC / DC hybrid grid based on risk analysis of long-term climate response capability of new energy. Figure 1 As shown in the figure, the execution process of the AC / DC hybrid grid planning method based on the risk analysis of the long-term climate response capability of new energy is as follows:

[0050] 1) Build a time series generation model based on a denoising diffusion transformer. Through the diffusion process and cross-scale attention mechanism, it captures the multi-timescale nonlinear coupling relationship between climate variables and renewable energy output, and generates a high-fidelity joint sample set.

[0051] 2) Extract the risk feature vector of the renewable energy output scenario and establish a mixed Gaussian distribution to represent the risk feature space;

[0052] 3) Based on the outlier detection algorithm, low-probability-high-impact scenarios are identified from the risk feature space to form a set of typical days for planning;

[0053] 4) Construct a mixed integer programming model with the goal of minimizing line construction costs, operating costs, and various penalty costs, covering all constraints such as AC / DC lines, thermal power / hydropower / new energy units;

[0054] 5) Solve the optimization model and obtain the AC / DC hybrid grid planning scheme.

[0055] The execution process and principle of the present invention are further described in detail below.

[0056] Exploring the long-term climate response patterns of renewable energy is a prerequisite for new power system planning methods. Its core lies in establishing a time-series generative model that accurately characterizes the nonlinear coupling relationship between renewable energy output and climate variables. Traditional planning methods typically use static probability distributions of historical meteorological data or typical scenario sets, failing to fully account for the spatiotemporal heterogeneity of renewable energy output caused by the long-term evolution of the climate system. This makes planning schemes unsuitable for multi-timescale risk assessment under the carbon neutrality goal. In particular, in scenarios with a high proportion of renewable energy grid integration, the output characteristics of clean energy sources such as wind power and photovoltaics are strongly dynamically correlated with climate factors such as wind speed, sunlight intensity, and temperature. Short-term fluctuations are dominated by transient changes in the weather system, while long-term trends are deeply tied to evolving climate patterns. This cross-timescale coupling significantly limits the generalization capabilities of traditional statistical modeling methods. Existing modeling methods rarely capture the high-dimensional nonlinear characteristics and long-term dependencies of climate factors simultaneously, resulting in significant deviations in renewable energy output simulation results at the interannual scale.

[0057] Based on the above situation, the present invention first constructs a time series generation model based on denoising diffusion Transformer. Through the diffusion process and cross-scale attention mechanism, it captures the multi-time-scale nonlinear coupling relationship between new energy output data and meteorological data in the planning area and generates a joint sample set.

[0058] In this step, the present invention constructs a time series generative model based on a denoising diffusion transformer, achieving an organic fusion of the multi-scale physical processes of the climate system and the probability distribution of renewable energy output, providing a high-fidelity, long-term climate-renewable energy joint sample set for subsequent risk analysis. In this time series generative model based on a denoising diffusion transformer, the progressive denoising mechanism of the diffusion model and the global attention mechanism of the transformer synergistically capture the cross-period correlation patterns of climate responses.

[0059] The diffusion process model of the climate-new energy coupling relationship is shown in equations (1)-(3):

[0060]

[0061] where q(·) represents the transition probability of the diffusion process; n and Ω N Respectively represent the index and set of diffusion steps, the total number of diffusion steps is recorded as N; X n and Y n represent the climate variables and renewable energy output sequence at the nth step in the diffusion process; Indicates the mean The covariance matrix is ​​β n The multivariate Gaussian distribution function of I is used in the diffusion model to describe the noise injection process from step n-1 to step n; β nis the noise scheduling coefficient of the nth step, satisfying β1<β2<…<β N ; Indicates the previous step data (X n-1 ,Y n-1 ) scaling; I is the unit matrix, indicating that each dimension is independently noised, and the corresponding β n I represents the diagonal covariance matrix, which means that the variance added independently in each dimension is β n Gaussian noise. X0 and Y0 represent the original climate variables and new energy output series respectively; x t and y t They represent the original climate variables and the new energy output values ​​at time t respectively; T is the number of sequence periods; d1 and d2 represent the vector dimensions of the original climate variables and the new energy output respectively; and denote the real vector spaces of d1 and d2 dimensions respectively.

[0062] The diffusion process of the climate-new energy coupling relationship constructs a Markov chain from the real data distribution to the isotropic Gaussian distribution, that is, how to perturb the data to Gaussian noise at each step. This design enables the subsequent inverse denoising process to reconstruct the original data distribution from pure noise by gradually removing the noise.

[0063] When constructing a long-term climate response model for renewable energy, the cross-scale attention mechanism is a key link connecting the diffusion process with physical laws. The denoising process is implemented through a conditional Transformer prediction network, the core of which is the multi-head attention (MHA) operation, which enables the collaborative modeling of multi-time-scale characteristics in the climate-new energy coupling relationship, as shown in Equations (4)-(5).

[0064] M(Q,K,V)=C(H1,H2,…,H h )W O (4)

[0065]

[0066] Where M(·) represents a multi-head attention operation; Q, K, and V represent the query matrix, key matrix, and value matrix, respectively; C(·) represents a vector concatenation function that concatenates the outputs of multiple attention heads into a high-dimensional matrix in the feature dimension; H i Represents the i-th independent attention head, which is the basic unit of multi-head parallel processing. It linearly transforms the input through an independent parameter matrix to generate queries, keys, and values ​​in the subspace; i and Ω head denotes the index and set of attention heads respectively; h is the number of attention heads; W Ois the output projection matrix; S(·) denotes a normalization exponential function for converting attention scores into probability distribution; and are the query, key and value projection matrices of the i-th attention head, respectively; d k denotes the key dimension, denotes the scaling factor. The core of the above modeling is to map the climate variable sequence X n and the new energy output sequence Y n into query matrix, key matrix and value matrix (in a specific embodiment, X n is mapped into query matrix and key matrix, and Y n is mapped into value matrix) through trainable projection weights and W i V perform feature transformation, and finally the correlation patterns of different subspaces are calculated in parallel by multiple attention heads, and the multi-head information is integrated by the output matrix to form the hidden state representation with physical meaning.

[0067] For the parameterization of the conditional denoising process, the inverse denoising process is defined as a parameterized Gaussian transition, and a neural network is used to learn how to remove noise step by step from the noisy data state X n and Y n to recover the original data and obtain the climate-new energy joint sample, as shown in equations (6)-(9):

[0068]

[0069]

[0070] where p θ denotes the transition probability of the inverse denoising process; μ θ (·) and Σ θ (·) are the mean and covariance functions parameterized by neural networks, respectively, for predicting the distribution of denoised data; α n denotes the cumulative noise attenuation coefficient considering the overall influence of the previous n steps of noise; h n is the hidden state generated by the cross-scale attention mechanism; P t is the time series position encoding; T(·) denotes the cross-scale attention mechanism; ε θ (·) denotes the Transformer noise prediction network; diag(·) denotes a function for converting a vector into a diagonal matrix; exp(·) denotes an exponential function; θ (·) denotes the log-variance prediction vector output by the Transformer network; β n is the noise scheduling coefficient of the n-th step; βs is the noise dispatch coefficient at step s. Through a parameterized denoising process, the model can gradually correct the noise data to approximate the true climate-renewable energy response while preserving dynamic characteristics across time scales. This allows the generation of corresponding renewable energy output curves based on future long-term climate scenarios.

[0071] The climate-new energy joint sample can be understood as multiple climate-new energy multi-dimensional time series curve samples in terms of specific expression. Each sample consists of (X t ,Y t ) and its corresponding distribution parameters. The corresponding distribution parameters include the climate and new energy data values ​​at the time series position, the characteristic dimensions of the sample climate variables and new energy output, the time series distribution mean and covariance of each sample data, and the decay weight of the joint sample with the number of steps in the diffusion-denoising process.

[0072] (2) Hierarchical screening of new energy output scenarios driven by risk sensitivity

[0073] Screening risk scenarios for the long-term climate responsiveness of renewable energy is a key step in AC / DC hybrid grid planning. Its goal is to extract representative scenarios with significant risk characteristics from a massive climate-renewable energy joint sample generated by a time-series generative model based on a denoising diffusion Transformer, providing a high-information-density input set for subsequent planning decisions. Traditional scenario screening methods, which rely on static clustering algorithms such as K-means, struggle to effectively identify low-probability-high-impact (LPHI) risk events caused by the nonlinear evolution of the climate system and have limited ability to capture spatiotemporal correlations. This method proposes a scenario screening framework driven by risk sensitivity to achieve refined extraction of risk scenarios.

[0074] Suppose the denoising diffusion transformer generates K new energy output scenarios through the transition probability of the reverse denoising process, and each scenario s k Corresponding to a risk feature vector, as shown in formula (10),

[0075]

[0076] Among them F k Represents scene s k The risk characteristic vector of k and Ω K Respectively represent the index and collection of new energy output scenarios; Represents scene s k The conditional risk value of D reflects the extreme loss expectation of renewable energy output in this scenario; k Represents scene s kThe spatiotemporal risk dispersion of S is used to measure the fluctuation difference of renewable energy output in different regions; k Represents scene s k The vulnerability entropy of T represents the nonlinear impact of climate variables on renewable energy output; k Represents scene s k The time scale coupling strength reflects the correlation between the short-term fluctuations and long-term trends of renewable energy output.

[0077] Specifically, the variables defined in formula (10) are as shown in formulas (11)-(17),

[0078]

[0079]

[0080] in It's a scene k Normalized output sequence of new energy stations under ; Represents scene s k The distribution function of the system power shortage under ; express α quantile of ; is the expected operation function; j represents the index of the new energy station; N Re Indicates the total number of new energy stations; For scenes k The normalized output of the new energy station j under ; For scenes k Average output of new energy stations under Me / l Me and Ω Me Represents the index and set of climate variables respectively; m Me is the climate variable dimension; and Represents scene s k Climate variables under Me and l Me The vulnerability coefficient is expressed as the partial derivative of formula (10); It's a scene k Climate variables under Me Normalized influence weight of n Wa / l Wa and Ω Wa are the index and set of wavelet decomposition scales respectively; N Wa is the total number of scales of wavelet decomposition; It's a scene k Under time scale n Wa Wavelet transform coefficients on ; is the scale n Wa energy share.

[0081] In order to screen out low probability and high impact LPHI scenarios, it is necessary to capture LPHI scenarios based on outlier detection. The goal is to extract the LPHI scenarios from the risk feature space F = {F1, F2, ..., F k} to identify outliers, i.e. LPHI scenario. Assume that the feature space obeys Q G A mixture of Gaussian distributions, as shown in formula (18),

[0082]

[0083] where q G is the index of the Gaussian distribution; For the qth G The mixing weight of the components is such that the sum of the components is 1; and The qth G The mean vector and covariance matrix of the components.

[0084] The outlier scene set is defined as the samples that fall outside the α-confidence interval of all Gaussian distributions, that is, the outlier point determination is shown in formula (19):

[0085]

[0086] Among them L LQ is the outlier scene set; 2 is the critical value of the chi-square distribution. Thus, the set of new energy output scenarios L that are judged as low probability-high impact scenarios is obtained. LQ .

[0087] (3) AC / DC hybrid network planning

[0088] The output scenario set obtained based on the risk analysis of the long-term climate response capacity of new energy is regarded as the typical day set within the planning target period. The typical day set includes the conventional climate-new energy scenario set sampled by the inverse transition probability of the diffusion process and the new energy output scenario set L that is judged to be a low probability-high impact scenario. LQ , where the new energy output scenario set L LQ It is considered as the core component of the typical day set. For the planning target year, the AC / DC hybrid power grid planning model constructed by this method aims to minimize the sum of line investment and construction costs, system operation costs under typical days, power shortage penalty costs, and power abandonment penalty costs. The objective function of the planning model is shown in formula (20):

[0089]

[0090] Among them C inv It is the sum of the investment and construction and system operation costs in the year; C ACinv and CDCinv are the investment and construction costs of AC and DC lines respectively; m / Ω rep and t / Ω T Represents the index / aggregation of typical day and intraday time periods respectively; p m represents the corresponding probability of a typical day m; and They represent the power shortage penalty cost, PV power curtailment penalty cost, wind power curtailment penalty cost, thermal power unit startup cost, thermal power unit shutdown cost, and thermal power unit operation cost in the t-th period of typical day m, respectively; represents the penalty cost of hydropower curtailment in a typical day m.

[0091] The line construction cost is expressed as formula (21)

[0092]

[0093] where j ACinv / Ω ACinv and j DCinv / Ω DCinv Respectively represent the index / set of the AC and DC lines to be selected; and They are candidate routes j ACinv and j DCinv The investment and construction cost; and They are respectively marked candidate line j ACinv and j DCinv Boolean variable indicating the construction status.

[0094] For the system operation part, it is expressed as formula (22)

[0095]

[0096] where i T / Ω Th Represents the index / collection of thermal power units; and They are thermal power units i T Unit startup, shutdown, and operating costs; and They are the marker thermal power unit i in the tth period of typical day m T Boolean variables for the startup and shutdown states; and represents the thermal power unit i in the tth period of a typical day m T Power generation.

[0097] For the penalty cost part, it is expressed as formula (23)

[0098]

[0099] where i / Ω B , i S / Ω S , i W / Ω W , and i H / Ω H denote the index / set of nodes, photovoltaic generators, wind power generators, and hydro power generators, respectively; v Shed , v SCur , v WCur , and v HCur denote the penalty coefficients of power shortage, photovoltaic curtailment, wind power curtailment, and hydro power curtailment, respectively; and denote the power shortage at node i B , the photovoltaic curtailment of generator i S , and the wind power curtailment of generator i W in the tth time interval of typical day m, respectively; is the hydro power curtailment of generator i H in typical day m.

[0100] The construction constraints of AC and DC lines are represented as shown in equations (24)-(27):

[0101]

[0102]

[0103] where and denote the transmission power of candidate AC line j ACinv and candidate DC line j DCinv in the tth time interval of typical day m, respectively; and are the maximum transmission power of candidate AC line j ACinv and candidate DC line j DCinv , respectively; T tot is the sum of the number of analysis time intervals in all typical days; μ ACinv and μ DCinv are the minimum utilization rates of AC and DC lines that meet the construction requirements, respectively.

[0104] For AC lines in the system, the power transmission characteristics are described by a DC power flow model, and the operation constraints are represented as shown in equations (28)-(29),

[0105]

[0106] where k / Ω AC and j ACex / Ω ACex Respectively represent the index / collection of the constructed AC lines and the original AC lines; and are the AC lines k and j in the tth period of a typical day m. ACex The transmission power; and are the voltage phase angles of the sending node and the receiving node corresponding to the AC line k in the t-th period of typical day m; is the AC line j ACex The maximum transmission power.

[0107] For the DC lines in the system, considering the complexity of their operation control mode, intraday power reverse transmission is not allowed during the modeling process, and their ramp performance is restricted, which can be expressed as shown in Equations (30)-(34).

[0108]

[0109] where j DCex / Ω DCex is the index / collection of the original DC lines; is the DC line j in the tth period of a typical day m DCex The transmission power; is the DC line j DCex Maximum transmission power; and They are respectively marked DC line j DCex and j DCinv Boolean variable for the direction of power transfer on a typical day m; and They are DC line j DCex and j DCinv Maximum climbing power; T d is the number of analysis periods within a single typical day.

[0110] The operating constraints of the thermal power units in the system are expressed as shown in equations (35)-(41).

[0111]

[0112] in is the index of thermal power unit i in the tth period of typical day m T Boolean variable of the running status; and They are thermal power units i T Minimum and maximum power generation; and They are thermal power units i T Minimum shutdown and startup times.

[0113] For the hydropower units in the system, a three-stage modeling scheme is adopted to constrain the maximum, minimum and average output power of the month in which the typical day is located. The operation constraints are expressed as shown in Equations (42)-(45):

[0114]

[0115]

[0116] in is the number of hydropower unit i in the tth period of a typical day m H Power generation capacity; It is a hydropower unit H installed capacity; and They are respectively hydropower unit i in the month corresponding to the typical day m H Maximum, minimum and average generated power.

[0117] The operating constraints of the new energy units in the system are expressed as shown in equations (46)-(49):

[0118]

[0119] in and are wind turbine i in the tth period of typical day m W and photovoltaic units S Power generation capacity; and are wind turbine i in the tth period of typical day m W

[0120] and photovoltaic units S Maximum power generation.

[0121] The power balance constraint in the system is expressed as shown in equations (50)-(51):

[0122]

[0123] in and are the indexes / sets of the original AC line, candidate AC line, original DC line, and candidate DC line with node i as the endpoint; and j S,i / Ω i S are the indexes / sets of thermal power units, hydropower units, wind power units and photovoltaic units at node i respectively; and They are the original AC line j with node i as the endpoint in the tth period of typical day m.ACex,i , candidate AC line j ACinv,i 、Original DC linej DCex,i and candidate DC line j DCinv,i The transmission power is negative when it is sent from node i and positive when it is received from node i; and They are the thermal power unit j at node i in the tth period of typical day m. T,i 、Hydrogen sets H,i 、wind turbines W,i and photovoltaic units S,i Power generation capacity; is the load demand at node i during the tth period of typical day m.

[0124] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0125] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for planning AC / DC hybrid grids based on climate responsiveness risk analysis, characterized in that: The following steps are involved: A time series generation model based on the denoising diffusion transformer is constructed. Through the diffusion process and cross-scale attention mechanism, the multi-time-scale nonlinear coupling relationship between renewable energy output data and meteorological data in the planning area is captured to generate a joint sample set. Extracting risk feature vectors of new energy output scenarios from the joint sample set and establishing a mixed Gaussian distribution to represent the risk feature space; Based on the outlier detection algorithm, low probability and high impact scenarios are identified from the risk feature space to form a set of typical days for planning; Based on the typical day set for planning, a mixed integer programming model is constructed with the goal of minimizing the line construction cost, operation cost and various penalty costs; The mixed integer programming model is solved to obtain an AC / DC hybrid grid planning scheme.

2. The AC / DC hybrid grid planning method based on climate responsiveness risk analysis according to claim 1 is characterized in that: The steps for building a time series generation model based on the denoising diffusion transformer include: Constructing a diffusion process model of climate-new energy coupling relationship: where q(·) represents the transition probability of the diffusion process; X0 and Y0 represent the original climate variables and the new energy output sequence, respectively; n and Ω N Respectively represent the index and set of diffusion steps, the total number of diffusion steps is recorded as N; X n and Y n represent the climate variables and renewable energy output sequence at the nth step in the diffusion process; Indicates the mean The covariance matrix is ​​β n The multivariate Gaussian distribution function of I is used to describe the noise injection process from step n-1 to step n; β n is the noise scheduling coefficient of the nth step, satisfying β1<β2<…<β N ; Indicates the previous step data (X n-1 ,Y n-1 ) scaling; I is the unit matrix, indicating that each dimension is independently noised, and the corresponding β n I represents the diagonal covariance matrix, which means that the variance added independently in each dimension is β n Gaussian noise; x t and y t They represent the original climate variables and the new energy output values ​​at time t respectively; T is the number of sequence periods; d1 and d2 represent the vector dimensions of the original climate variables and the new energy output respectively; R d1 and R d2 denote the real vector space of d1 dimension and d2 dimension respectively; The climate variable sequence X injected with diffusion noise n Mapped to the query matrix and key matrix of the multi-head attention mechanism in the Transformer neural network, the new energy output sequence Y injected with noise during the diffusion process is transformed into n Mapped to the value matrix of the multi-head attention mechanism in the Transformer neural network, and through the trainable projection weights and Feature transformation is performed, and finally multiple attention heads calculate the correlation patterns of different subspaces in parallel. The output matrix integrates the multi-head information to form a hidden state representation with physical meaning.

3. The AC / DC hybrid grid planning method based on climate responsiveness risk analysis according to claim 2, characterized in that: Capturing the multi-time-scale nonlinear coupling relationship between renewable energy output data and meteorological data in the planning area and generating a joint sample set involves the following steps: The hidden state representation with physical meaning is gradually de-noised through the inverse process, and the multi-time-scale nonlinear coupling relationship between the renewable energy output data and meteorological data in the planning area is captured.

4. The AC / DC hybrid grid planning method based on climate responsiveness risk analysis according to claim 3 is characterized in that: The hidden state representation with physical meaning is gradually de-noised through the reverse process, which is specifically achieved through the following formula: where p θ represents the transition probability of the inverse denoising process; μ θ (·) and Σ θ (·) are the mean and covariance functions parameterized by the Transformer neural network, respectively, used to predict the distribution of denoised data; α n It represents the cumulative noise attenuation coefficient considering the overall influence of the noise in the previous n steps; h n is the hidden state generated by the cross-scale attention mechanism; P t is the temporal position encoding; T(·) represents the cross-scale attention mechanism; ε θ (·) represents the Transformer noise prediction network; diag(·) represents the function of converting a vector into a diagonal matrix; exp(·) represents the exponential function; represents the logarithmic variance prediction vector of the conditional Transformer network output; β n is the noise scheduling coefficient of the nth step; β s is the noise scheduling coefficient of the sth step.

5. The AC / DC hybrid grid planning method based on climate responsiveness risk analysis according to claim 1, characterized in that: The risk feature vectors of the new energy output scenarios are extracted from the joint sample set, and a mixed Gaussian distribution is established to represent the risk feature space, specifically including: The denoising diffusion transformer temporal generation model generates K renewable energy output scenarios through the transition probability of the reverse denoising process; Extract risk feature vectors from the K new energy output scenarios, and each scenario s k Corresponding to a risk feature vector, a mixed Gaussian distribution is established based on the risk feature vector to represent the risk feature space: Specifically, the variables defined in formula (10) are shown in formulas (11)-(17): in It's a scene k Normalized output sequence of new energy stations under ; Represents scene s k The distribution function of the system power shortage under ; express α quantile; E[·] is the expected operation function; j represents the index of the new energy station; N Re Indicates the total number of new energy stations; For scenes k The normalized output of the new energy station j under ; For scenes k Average output of new energy stations under Me / l Me and Ω Me Represents the index and set of climate variables respectively; m Me is the climate variable dimension; and Represents scene s k Climate variables under Me and l Me The vulnerability coefficient is expressed as the partial derivative of formula (10); It's a scene k Climate variables under Me Normalized influence weight of n Wa / l Wa and Ω Wa are the index and set of wavelet decomposition scales respectively; N Wa is the total number of scales of wavelet decomposition; It's a scene k Under time scale n Wa Wavelet transform coefficients on ; is the scale n Wa energy share.

6. The AC / DC hybrid grid planning method based on climate responsiveness risk analysis according to claim 1, characterized in that: Based on the outlier detection algorithm, low-probability-high-impact scenarios are identified from the risk feature space to form a set of typical days for planning, specifically including: Assume that the feature space obeys Q G A mixture of Gaussian distributions, as shown in formula (18), where q G is the index of the Gaussian distribution; For the qth G The mixing weight of the components is such that the sum of the components is 1; and The qth G The mean vector and covariance matrix of the components; The outlier scene set is defined as the samples that fall outside the α-confidence interval of all Gaussian distributions, that is, the outlier point determination is shown in formula (19): Among them L LQ is the outlier scene set; 2 is the critical value of the chi-square distribution, and thus the set of new energy output scenarios that are judged to be low probability-high impact scenarios is obtained; The new energy output scenario set of low probability-high impact scenario and the conventional climate-new energy scenario set sampled by the inverse transition probability of the diffusion process are combined to form the typical day set for planning.

7. The AC / DC hybrid grid planning method based on climate responsiveness risk analysis according to claim 1, characterized in that: Based on the typical day set for planning, a mixed integer programming model is constructed with the goal of minimizing the line construction cost, operating cost, and various penalty costs. Specifically, the model includes: Among them C inv It is the sum of the investment and construction and system operation costs in the year; C ACinv and C DCinv are the investment and construction costs of AC and DC lines respectively; m / Ω rep and t / Ω T Represents the index / aggregation of typical day and intraday time periods respectively; p m represents the corresponding probability of a typical day m; and They represent the power shortage penalty cost, PV power curtailment penalty cost, wind power curtailment penalty cost, thermal power unit startup cost, thermal power unit shutdown cost, and thermal power unit operation cost in the t-th period of typical day m, respectively; represents the penalty cost of hydropower curtailment in a typical day m.

Images